wording changes from chat

lw3/report.md

@@ -162,10 +162,10 @@ This configuration allows SVM to model non-linear class boundaries.
 
 - **Seeds Dataset**
   - Accuracy $\approx$ **87.3%**
-  - Lower than k-NN, likely due to suboptimal hyperparameters or the specific class structure
+  - Lower than k-NN, likely influenced by class overlap and the relatively small sample size, which limits the margin-based advantages of SVM in this setting
 
 **Conclusion:**
-SVM achieves perfect separation on linearly separable data (Banknote) but does not always outperform k-NN. Performance depends on hyperparameter tuning and dataset characteristics.
+SVM achieves perfect separation on linearly separable data (Banknote) but does not always outperform k-NN. Performance depends on dataset characteristics such as class separability and sample size.
 
 ## 2.4 Multi-class handling in SVM